Application of improved bacterial foraging optimization in Bayesian network structure learning
JIN Tong, LIN Feng
(School of Electrical Engineering, Zhejiang University, Hangzhou 310027, China)

Abstract: Aiming at the problems that the existing Bayesian network structure learning algorithms always had indequacies such as easy to premature convergence, low learning efficiency, a Bayesian network structure learning strategy based on improved bacterial foraging optimization algorithm was proposed, which improved the chemotaxis operator, reproduction operator and elimination operator in the traditional bacterial foraging optimization. The adaptive theory was applied to the calculation of bacterial run step size and selection of reproduction individuals. In the migration probability calculation of the migration operator, the roulette method in the genetic algorithm was introduced. Based on mutual information theory, a new random evolution method of the network structure was created to take place of the random elimination method in the traditional bacterial optimization. Experiments on classical Bayesian networks of different scales were proceeded. The results indicate that the algorithm is effective in Bayesian network structure learning, the performance in convergence is inferior to other algorithms, but in the learning effect, especially for the complex structure of the network, the advantages are obvious.

Key words: Bayesian networks; bacterial foraging optimization (BFO); structure learning; mutual information theory